The editor of Downcodes learned that ByteDance’s Doubao model team has released SeedEdit, China’s first commercialized universal image editing model, and has started testing it on Doubao’s PC side and Jimeng’s web side. SeedEdit allows users to achieve professional-level image editing effects with just simple text commands, and its technology is a breakthrough. The model supports Chinese and English input, and can accurately identify Chinese idioms and proper nouns. It performs well in benchmark evaluations such as HQ-Edit, especially when processing images generated by the model, which is better than real images.
In performance tests, SeedEdit performed well on benchmarks such as HQ-Edit, not only being able to understand fuzzy instructions but also having higher image retention and success rates when performing fine edits. Currently, the model supports Chinese and English input, and can accurately identify Chinese idioms and proper nouns.
The Doubao team stated that although SeedEdit currently performs better than real images when processing images generated by the model, the team is continuing to optimize model performance. In the future, we will focus on improving the real image processing effect, ID retention capability and editing accuracy, and explore the generation of long-term story and comic content to provide users with more powerful creative tools.
Experience address: https://top.aibase.com/tool/seededit
The emergence of SeedEdit marks a significant progress in domestic AI image editing technology, providing users with a more convenient and efficient way to edit images. The editor of Downcodes hopes that SeedEdit can further improve the real image processing effect in the future and expand more application scenarios to bring users a more amazing creative experience.